JavaScript's sweet spot

http://blog.mrale.ph/post/12396216081/the-trap-of-the-performance-sweet-spot

by paufernandez • 15 years ago

115 18 15 years ago

blog.mrale.ph

christkv2• 15 years ago

awesome article and a good balance between the "possible" speed of js vs the sacrifice you make in code quality to get that speed.

mraleph5• 15 years ago

Well I would not recommend writing code like mr. C to anyone. Unless you really-really-really need every last usec you can get :-)

The goal was just to show that sometimes JS code generated by some tool sometimes can be faster than handwritten JS code just because it tries to emulate something foreign (like low-level memory semantics) and thus reducing certain overheads. So one should be very careful when extrapolating speed shown by some project into effectiveness of the whole language.

[also I tried to stress the fact that speed is not everything there is to performance. If you have a long lived webapp in your tab you suddenly might become concerned by memory usage. Which is rather difficult to predict partially because of the "hairy" nature of the heap, the fact that VM tries to adapt to your app but doesn't always get it right, and the cost of the VM itself (e.g. generated code and additional data structures)]

christkv2• 15 years ago

but for some very narrow hotspots in your code it can mean a massive difference :)

mbq• 15 years ago

The worse is that people expect such language to perform dumb high-load operations rather than use its expressiveness to conduct chunks of lower-level language to do the dirty job, such as with <canvas> over SVG. R is from an another world, but it is a perfect example how this idea can work very well.

Srirangan2• 15 years ago

>> Suddenly mr. C gets a call from his boss. “We’ve decided >> to rewrite our apps in JavaScript. It is pretty fast >> nowdays, ya know.” says the boss and hangs up.

Umm... Instead of proceeding to rewrite the app and author this blog, the author could have realized that JavaScript (or Scala, Java, Python etc.) could realistically be faster than C and should have explained the same to his/her boss.

Srirangan2• 15 years ago

Typo: could realistically not be faster

mraleph5• 15 years ago

Well... You are underestimating the power of sweet spot :-) In a tight computation kernel it's actually possible for JavaScript or any other JIT compiled language given a good JIT to match the performance of C.

Also mr. C and his boss came into existence for a very particular purpose and that purpose is not "JavaScript vs. C" but rather "Normal JavaScript vs. inhuman JavaScript" comparison.

masklinn• 15 years ago

The boss did not say javascript was as fast as C, but that it was "pretty fast". As in "fast enough", with the advantage of simpler distribution.

And I'd tend to agree, a single order of magnitude difference with C out of the box? That's pretty damn good (it helps that the code is highly numerical, JITs are good at that).

Sigma0• 15 years ago

This is an interesting article with regards to helping the V8 compiler optimise JavaScript code. Often, if C like speed is required, it would be easier to just write C rather than try emulate it through esoteric language features, but I wonder if these kind of compiler optimisations could be included via a language that compiles to JavaScript, such as CoffeeScript, at the expense of readability.

mraleph5• 15 years ago

emscripten generates very similar code from LLVM bitcode which in turn was generated (optimized) from say C or C++.

It's next to impossible to emit such code from language that is not statically typed.

Also it's not necessarily going to be faster than handwritten "human" JavaScript. There are a lot of factors in play. That's exactly what I am trying to stress in the blog post.

masterj• 15 years ago

I think the OP was suggesting that another language could be used for this purpose, similar in function to CoffeeScript in that it compiles down to JavaScript, but statically typed (or even just straight C). That way most of your app could be written in CoffeeScript or JavaScript, but for those few functions who's performance REALLY matters, you could reach for a different tool that makes writing this inhumane but performant flavor of JavaScript easy.

It's an interesting idea, and having little to no knowledge of compilers, I'd be interested in your take on it. Could emscripten already be used for something like this?

Peaker• 15 years ago

Good article. I strongly disagree with this though:

> And implementation complexity is the worst kind of complexity in the world: it leads to subtle bugs that occur on 0.000001% of programs, it leads to unpredictable and hard to analyze performance, it leads to volumes of rulebooks that specify how to appease the compiler

Implementation complexity is much better than interface complexity. It does not necessarily break down subtly. Unlike interface complexity, it is localized.

scubaguy• 15 years ago

Android (VM based language) has the following API:

public View getDropDownView (int position, View convertView, ViewGroup parent)

convertView - the old view to reuse, if possible. Note: You should check that this view is non-null and of an appropriate type before using. If it is not possible to convert this view to display the correct data, this method can create a new view.

The presence of the unintuitive 'convertView' object is caused by the underlying VM and the unpredictable GC nature. This demonstrates that implementation complexity can lead to a complex interface.

IMO the design of the interface is an implementation. A language that is complex to implement will lead to complex interfaces.

wickedchicken2• 15 years ago

is the author a C programmer or a C++ programmer? naming a struct 'Point' is suspicious (instead of 'point' or 'POINT')

mraleph5• 15 years ago

yeah, I am sorry, I did not follow C naming etiquette to the letter. You are right: these days I am a bit accustomed to starting struct's names with a capital letter because I do a lot of C++ programming.

(I would not call the struct POINT though because it would feel like something from WinAPI :-))

wickedchicken2• 15 years ago

ender7• 15 years ago

What kind of performance increases could we see if we simply added type annotations to Javascript? Would this give the VMs a significant boost?

Something like AS3 (Harmony) syntax:

  var foo:int = 3;
  var bar:string = "A nice string";
  var baz:Array = ["a", "polymorphous", 4, "array"];
  var v:Array.<int> = [2,3,4];

  function f(a:int, b:string):string {
    return a + " " + b;
  }

Sidenote: I really want someone to write a Coffeescript compiler that supports type annotations like the above. It can strip them out when it compiles the code; I just want it to yell at me if I violate one of my declared types.

mraleph5• 15 years ago

It really depends on the semantics of those annotations.

For example: how is v:Array.<int> enforced (and is it enforced at all)?

Type annotations can be a good to seed type inference analysis or adjust collected type feedback. But unless they guarantee stability of objects shapes (e.g. Array.<float> can contain only floating point numbers) they'll give little.

JavaScript's sweet spot

Related Stories